Thanks, Sunil! I found your JIRA (YARN-1963 <https://issues.apache.org/jira/browse/YARN-1963>) that has a really great design doc.
On Mon, Nov 13, 2017 at 6:04 PM, Sunil G <[email protected]> wrote: > Hi Benson, > > Prior to 2.8 releases, YARN did not support priorities for its > applications. Currently user can specify priority (higher integer value > means higher priority) to its applications so that high priority apps could > get resources faster from scheduler (priority is applicable within a leaf > queue). > > In case of MR, we kept the same labels though. And each of this label is > mapped to an integer like VERY_HIGH : 5; HIGH : 4; NORMAL : 3; LOW : 2; > VERY_LOW : 1; DEFAULT : 0. > > Apart from this labels, one can also specify an integer value to > mapreduce.job.priority property. > This means that mapreduce.job.priority = HIGH & mapreduce.job.priority = > 7 are valid. > > - Sunil > > > On Tue, Nov 14, 2017 at 4:22 AM Benson Qiu <[email protected]> > wrote: > >> I'm having trouble finding documentation for JobPriority >> (mapreduce.job.priority). >> >> > "Changes the priority of the job. Allowed priority values are >> VERY_HIGH, HIGH, NORMAL, LOW, VERY_LOW" >> >> I am using YARN CapacityScheduler, with preemption disabled. What exactly >> does "low priority" or "high priority" mean in terms of how containers are >> allocated to jobs? >> >> Thanks, >> Benson >> >
